4 * Copyright (c) 2016 Red Hat, Inc.
7 * Victor Kaplansky <victork@redhat.com>
8 * Marc-André Lureau <mlureau@redhat.com>
10 * This work is licensed under the terms of the GNU GPL, version 2 or
11 * later. See the COPYING file in the top-level directory.
14 #ifndef LIBVHOST_USER_H
15 #define LIBVHOST_USER_H
21 #include <linux/vhost.h>
22 #include "standard-headers/linux/virtio_ring.h"
24 /* Based on qemu/hw/virtio/vhost-user.c */
25 #define VHOST_USER_F_PROTOCOL_FEATURES 30
26 #define VHOST_LOG_PAGE 4096
28 #define VHOST_MAX_NR_VIRTQUEUE 8
29 #define VIRTQUEUE_MAX_SIZE 1024
31 #define VHOST_MEMORY_MAX_NREGIONS 8
33 typedef enum VhostSetConfigType
{
34 VHOST_SET_CONFIG_TYPE_MASTER
= 0,
35 VHOST_SET_CONFIG_TYPE_MIGRATION
= 1,
39 * Maximum size of virtio device config space
41 #define VHOST_USER_MAX_CONFIG_SIZE 256
43 enum VhostUserProtocolFeature
{
44 VHOST_USER_PROTOCOL_F_MQ
= 0,
45 VHOST_USER_PROTOCOL_F_LOG_SHMFD
= 1,
46 VHOST_USER_PROTOCOL_F_RARP
= 2,
47 VHOST_USER_PROTOCOL_F_REPLY_ACK
= 3,
48 VHOST_USER_PROTOCOL_F_NET_MTU
= 4,
49 VHOST_USER_PROTOCOL_F_SLAVE_REQ
= 5,
50 VHOST_USER_PROTOCOL_F_CROSS_ENDIAN
= 6,
51 VHOST_USER_PROTOCOL_F_CRYPTO_SESSION
= 7,
52 VHOST_USER_PROTOCOL_F_PAGEFAULT
= 8,
53 VHOST_USER_PROTOCOL_F_CONFIG
= 9,
54 VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD
= 10,
55 VHOST_USER_PROTOCOL_F_HOST_NOTIFIER
= 11,
56 VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
= 12,
58 VHOST_USER_PROTOCOL_F_MAX
61 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
63 typedef enum VhostUserRequest
{
65 VHOST_USER_GET_FEATURES
= 1,
66 VHOST_USER_SET_FEATURES
= 2,
67 VHOST_USER_SET_OWNER
= 3,
68 VHOST_USER_RESET_OWNER
= 4,
69 VHOST_USER_SET_MEM_TABLE
= 5,
70 VHOST_USER_SET_LOG_BASE
= 6,
71 VHOST_USER_SET_LOG_FD
= 7,
72 VHOST_USER_SET_VRING_NUM
= 8,
73 VHOST_USER_SET_VRING_ADDR
= 9,
74 VHOST_USER_SET_VRING_BASE
= 10,
75 VHOST_USER_GET_VRING_BASE
= 11,
76 VHOST_USER_SET_VRING_KICK
= 12,
77 VHOST_USER_SET_VRING_CALL
= 13,
78 VHOST_USER_SET_VRING_ERR
= 14,
79 VHOST_USER_GET_PROTOCOL_FEATURES
= 15,
80 VHOST_USER_SET_PROTOCOL_FEATURES
= 16,
81 VHOST_USER_GET_QUEUE_NUM
= 17,
82 VHOST_USER_SET_VRING_ENABLE
= 18,
83 VHOST_USER_SEND_RARP
= 19,
84 VHOST_USER_NET_SET_MTU
= 20,
85 VHOST_USER_SET_SLAVE_REQ_FD
= 21,
86 VHOST_USER_IOTLB_MSG
= 22,
87 VHOST_USER_SET_VRING_ENDIAN
= 23,
88 VHOST_USER_GET_CONFIG
= 24,
89 VHOST_USER_SET_CONFIG
= 25,
90 VHOST_USER_CREATE_CRYPTO_SESSION
= 26,
91 VHOST_USER_CLOSE_CRYPTO_SESSION
= 27,
92 VHOST_USER_POSTCOPY_ADVISE
= 28,
93 VHOST_USER_POSTCOPY_LISTEN
= 29,
94 VHOST_USER_POSTCOPY_END
= 30,
95 VHOST_USER_GET_INFLIGHT_FD
= 31,
96 VHOST_USER_SET_INFLIGHT_FD
= 32,
100 typedef enum VhostUserSlaveRequest
{
101 VHOST_USER_SLAVE_NONE
= 0,
102 VHOST_USER_SLAVE_IOTLB_MSG
= 1,
103 VHOST_USER_SLAVE_CONFIG_CHANGE_MSG
= 2,
104 VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG
= 3,
106 } VhostUserSlaveRequest
;
108 typedef struct VhostUserMemoryRegion
{
109 uint64_t guest_phys_addr
;
110 uint64_t memory_size
;
111 uint64_t userspace_addr
;
112 uint64_t mmap_offset
;
113 } VhostUserMemoryRegion
;
115 typedef struct VhostUserMemory
{
118 VhostUserMemoryRegion regions
[VHOST_MEMORY_MAX_NREGIONS
];
121 typedef struct VhostUserLog
{
123 uint64_t mmap_offset
;
126 typedef struct VhostUserConfig
{
130 uint8_t region
[VHOST_USER_MAX_CONFIG_SIZE
];
133 static VhostUserConfig c
__attribute__ ((unused
));
134 #define VHOST_USER_CONFIG_HDR_SIZE (sizeof(c.offset) \
138 typedef struct VhostUserVringArea
{
142 } VhostUserVringArea
;
144 typedef struct VhostUserInflight
{
146 uint64_t mmap_offset
;
152 # define VU_PACKED __attribute__((gcc_struct, packed))
154 # define VU_PACKED __attribute__((packed))
157 typedef struct VhostUserMsg
{
160 #define VHOST_USER_VERSION_MASK (0x3)
161 #define VHOST_USER_REPLY_MASK (0x1 << 2)
162 #define VHOST_USER_NEED_REPLY_MASK (0x1 << 3)
164 uint32_t size
; /* the following payload size */
167 #define VHOST_USER_VRING_IDX_MASK (0xff)
168 #define VHOST_USER_VRING_NOFD_MASK (0x1 << 8)
170 struct vhost_vring_state state
;
171 struct vhost_vring_addr addr
;
172 VhostUserMemory memory
;
174 VhostUserConfig config
;
175 VhostUserVringArea area
;
176 VhostUserInflight inflight
;
179 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
182 } VU_PACKED VhostUserMsg
;
184 typedef struct VuDevRegion
{
185 /* Guest Physical address. */
187 /* Memory region size. */
189 /* QEMU virtual address (userspace). */
191 /* Starting offset in our mmaped space. */
192 uint64_t mmap_offset
;
193 /* Start address of mmaped space. */
197 typedef struct VuDev VuDev
;
199 typedef uint64_t (*vu_get_features_cb
) (VuDev
*dev
);
200 typedef void (*vu_set_features_cb
) (VuDev
*dev
, uint64_t features
);
201 typedef int (*vu_process_msg_cb
) (VuDev
*dev
, VhostUserMsg
*vmsg
,
203 typedef void (*vu_queue_set_started_cb
) (VuDev
*dev
, int qidx
, bool started
);
204 typedef bool (*vu_queue_is_processed_in_order_cb
) (VuDev
*dev
, int qidx
);
205 typedef int (*vu_get_config_cb
) (VuDev
*dev
, uint8_t *config
, uint32_t len
);
206 typedef int (*vu_set_config_cb
) (VuDev
*dev
, const uint8_t *data
,
207 uint32_t offset
, uint32_t size
,
210 typedef struct VuDevIface
{
211 /* called by VHOST_USER_GET_FEATURES to get the features bitmask */
212 vu_get_features_cb get_features
;
213 /* enable vhost implementation features */
214 vu_set_features_cb set_features
;
215 /* get the protocol feature bitmask from the underlying vhost
217 vu_get_features_cb get_protocol_features
;
218 /* enable protocol features in the underlying vhost implementation. */
219 vu_set_features_cb set_protocol_features
;
220 /* process_msg is called for each vhost-user message received */
221 /* skip libvhost-user processing if return value != 0 */
222 vu_process_msg_cb process_msg
;
223 /* tells when queues can be processed */
224 vu_queue_set_started_cb queue_set_started
;
226 * If the queue is processed in order, in which case it will be
227 * resumed to vring.used->idx. This can help to support resuming
228 * on unmanaged exit/crash.
230 vu_queue_is_processed_in_order_cb queue_is_processed_in_order
;
231 /* get the config space of the device */
232 vu_get_config_cb get_config
;
233 /* set the config space of the device */
234 vu_set_config_cb set_config
;
237 typedef void (*vu_queue_handler_cb
) (VuDev
*dev
, int qidx
);
239 typedef struct VuRing
{
241 struct vring_desc
*desc
;
242 struct vring_avail
*avail
;
243 struct vring_used
*used
;
244 uint64_t log_guest_addr
;
248 typedef struct VuDescStateSplit
{
249 /* Indicate whether this descriptor is inflight or not.
250 * Only available for head-descriptor. */
256 /* Maintain a list for the last batch of used descriptors.
257 * Only available when batching is used for submitting */
260 /* Used to preserve the order of fetching available descriptors.
261 * Only available for head-descriptor. */
265 typedef struct VuVirtqInflight
{
266 /* The feature flags of this region. Now it's initialized to 0. */
269 /* The version of this region. It's 1 currently.
270 * Zero value indicates a vm reset happened. */
273 /* The size of VuDescStateSplit array. It's equal to the virtqueue
274 * size. Slave could get it from queue size field of VhostUserInflight. */
277 /* The head of list that track the last batch of used descriptors. */
278 uint16_t last_batch_head
;
280 /* Storing the idx value of used ring */
283 /* Used to track the state of each descriptor in descriptor table */
284 VuDescStateSplit desc
[0];
287 typedef struct VuVirtqInflightDesc
{
290 } VuVirtqInflightDesc
;
292 typedef struct VuVirtq
{
295 VuVirtqInflight
*inflight
;
297 VuVirtqInflightDesc
*resubmit_list
;
299 uint16_t resubmit_num
;
303 /* Next head to pop */
304 uint16_t last_avail_idx
;
306 /* Last avail_idx read from VQ. */
307 uint16_t shadow_avail_idx
;
311 /* Last used index value we have signalled on */
312 uint16_t signalled_used
;
314 /* Last used index value we have signalled on */
315 bool signalled_used_valid
;
317 /* Notification enabled? */
322 vu_queue_handler_cb handler
;
331 enum VuWatchCondtion
{
332 VU_WATCH_IN
= POLLIN
,
333 VU_WATCH_OUT
= POLLOUT
,
334 VU_WATCH_PRI
= POLLPRI
,
335 VU_WATCH_ERR
= POLLERR
,
336 VU_WATCH_HUP
= POLLHUP
,
339 typedef void (*vu_panic_cb
) (VuDev
*dev
, const char *err
);
340 typedef void (*vu_watch_cb
) (VuDev
*dev
, int condition
, void *data
);
341 typedef void (*vu_set_watch_cb
) (VuDev
*dev
, int fd
, int condition
,
342 vu_watch_cb cb
, void *data
);
343 typedef void (*vu_remove_watch_cb
) (VuDev
*dev
, int fd
);
345 typedef struct VuDevInflightInfo
{
354 VuDevRegion regions
[VHOST_MEMORY_MAX_NREGIONS
];
355 VuVirtq vq
[VHOST_MAX_NR_VIRTQUEUE
];
356 VuDevInflightInfo inflight_info
;
362 uint64_t protocol_features
;
365 /* @set_watch: add or update the given fd to the watch set,
366 * call cb when condition is met */
367 vu_set_watch_cb set_watch
;
369 /* @remove_watch: remove the given fd from the watch set */
370 vu_remove_watch_cb remove_watch
;
372 /* @panic: encountered an unrecoverable error, you may try to
375 const VuDevIface
*iface
;
379 bool postcopy_listening
;
382 typedef struct VuVirtqElement
{
384 unsigned int out_num
;
387 struct iovec
*out_sg
;
392 * @dev: a VuDev context
393 * @socket: the socket connected to vhost-user master
394 * @panic: a panic callback
395 * @set_watch: a set_watch callback
396 * @remove_watch: a remove_watch callback
397 * @iface: a VuDevIface structure with vhost-user device callbacks
399 * Intializes a VuDev vhost-user context.
401 void vu_init(VuDev
*dev
,
404 vu_set_watch_cb set_watch
,
405 vu_remove_watch_cb remove_watch
,
406 const VuDevIface
*iface
);
411 * @dev: a VuDev context
413 * Cleans up the VuDev context
415 void vu_deinit(VuDev
*dev
);
419 * @dev: a VuDev context
421 * Process one vhost-user message.
423 * Returns: TRUE on success, FALSE on failure.
425 bool vu_dispatch(VuDev
*dev
);
429 * @dev: a VuDev context
430 * @plen: guest memory size
431 * @guest_addr: guest address
433 * Translate a guest address to a pointer. Returns NULL on failure.
435 void *vu_gpa_to_va(VuDev
*dev
, uint64_t *plen
, uint64_t guest_addr
);
439 * @dev: a VuDev context
442 * Returns the queue number @qidx.
444 VuVirtq
*vu_get_queue(VuDev
*dev
, int qidx
);
447 * vu_set_queue_handler:
448 * @dev: a VuDev context
449 * @vq: a VuVirtq queue
450 * @handler: the queue handler callback
452 * Set the queue handler. This function may be called several times
453 * for the same queue. If called with NULL @handler, the handler is
456 void vu_set_queue_handler(VuDev
*dev
, VuVirtq
*vq
,
457 vu_queue_handler_cb handler
);
460 * vu_set_queue_host_notifier:
461 * @dev: a VuDev context
462 * @vq: a VuVirtq queue
463 * @fd: a file descriptor
464 * @size: host page size
465 * @offset: notifier offset in @fd file
467 * Set queue's host notifier. This function may be called several
468 * times for the same queue. If called with -1 @fd, the notifier
471 bool vu_set_queue_host_notifier(VuDev
*dev
, VuVirtq
*vq
, int fd
,
472 int size
, int offset
);
475 * vu_queue_set_notification:
476 * @dev: a VuDev context
477 * @vq: a VuVirtq queue
480 * Set whether the queue notifies (via event index or interrupt)
482 void vu_queue_set_notification(VuDev
*dev
, VuVirtq
*vq
, int enable
);
486 * @dev: a VuDev context
487 * @vq: a VuVirtq queue
489 * Returns: whether the queue is enabled.
491 bool vu_queue_enabled(VuDev
*dev
, VuVirtq
*vq
);
495 * @dev: a VuDev context
496 * @vq: a VuVirtq queue
498 * Returns: whether the queue is started.
500 bool vu_queue_started(const VuDev
*dev
, const VuVirtq
*vq
);
504 * @dev: a VuDev context
505 * @vq: a VuVirtq queue
507 * Returns: true if the queue is empty or not ready.
509 bool vu_queue_empty(VuDev
*dev
, VuVirtq
*vq
);
513 * @dev: a VuDev context
514 * @vq: a VuVirtq queue
516 * Request to notify the queue via callfd (skipped if unnecessary)
518 void vu_queue_notify(VuDev
*dev
, VuVirtq
*vq
);
522 * @dev: a VuDev context
523 * @vq: a VuVirtq queue
524 * @sz: the size of struct to return (must be >= VuVirtqElement)
526 * Returns: a VuVirtqElement filled from the queue or NULL. The
527 * returned element must be free()-d by the caller.
529 void *vu_queue_pop(VuDev
*dev
, VuVirtq
*vq
, size_t sz
);
534 * @dev: a VuDev context
535 * @vq: a VuVirtq queue
536 * @elem: The #VuVirtqElement
537 * @len: number of bytes written
539 * Pretend the most recent element wasn't popped from the virtqueue. The next
540 * call to vu_queue_pop() will refetch the element.
542 void vu_queue_unpop(VuDev
*dev
, VuVirtq
*vq
, VuVirtqElement
*elem
,
547 * @dev: a VuDev context
548 * @vq: a VuVirtq queue
549 * @num: number of elements to push back
551 * Pretend that elements weren't popped from the virtqueue. The next
552 * virtqueue_pop() will refetch the oldest element.
554 * Returns: true on success, false if @num is greater than the number of in use
557 bool vu_queue_rewind(VuDev
*dev
, VuVirtq
*vq
, unsigned int num
);
561 * @dev: a VuDev context
562 * @vq: a VuVirtq queue
563 * @elem: a VuVirtqElement
564 * @len: length in bytes to write
565 * @idx: optional offset for the used ring index (0 in general)
567 * Fill the used ring with @elem element.
569 void vu_queue_fill(VuDev
*dev
, VuVirtq
*vq
,
570 const VuVirtqElement
*elem
,
571 unsigned int len
, unsigned int idx
);
575 * @dev: a VuDev context
576 * @vq: a VuVirtq queue
577 * @elem: a VuVirtqElement
578 * @len: length in bytes to write
580 * Helper that combines vu_queue_fill() with a vu_queue_flush().
582 void vu_queue_push(VuDev
*dev
, VuVirtq
*vq
,
583 const VuVirtqElement
*elem
, unsigned int len
);
587 * @dev: a VuDev context
588 * @vq: a VuVirtq queue
589 * @num: number of elements to flush
591 * Mark the last number of elements as done (used.idx is updated by
594 void vu_queue_flush(VuDev
*dev
, VuVirtq
*vq
, unsigned int num
);
597 * vu_queue_get_avail_bytes:
598 * @dev: a VuDev context
599 * @vq: a VuVirtq queue
600 * @in_bytes: in bytes
601 * @out_bytes: out bytes
602 * @max_in_bytes: stop counting after max_in_bytes
603 * @max_out_bytes: stop counting after max_out_bytes
605 * Count the number of available bytes, up to max_in_bytes/max_out_bytes.
607 void vu_queue_get_avail_bytes(VuDev
*vdev
, VuVirtq
*vq
, unsigned int *in_bytes
,
608 unsigned int *out_bytes
,
609 unsigned max_in_bytes
, unsigned max_out_bytes
);
612 * vu_queue_avail_bytes:
613 * @dev: a VuDev context
614 * @vq: a VuVirtq queue
615 * @in_bytes: expected in bytes
616 * @out_bytes: expected out bytes
618 * Returns: true if in_bytes <= in_total && out_bytes <= out_total
620 bool vu_queue_avail_bytes(VuDev
*dev
, VuVirtq
*vq
, unsigned int in_bytes
,
621 unsigned int out_bytes
);
623 #endif /* LIBVHOST_USER_H */